Every team I know has one sentence that is meant to slow people down.
For us, it was this: keep a human in the loop.
We said it in design reviews. We said it before a release. We said it whenever a tool produced an answer that looked confident, clean, and just a little too fast to trust.
Then one day somebody said it as a joke, right after an automated check approved something it should never have approved. The room laughed, because we all knew the real story. The system had done exactly what we asked it to do. The problem was what we asked.
That is where this design started.
The phrase earns its keep
"Human in the Loop" is one of the COMPILE & COTTON >_ designs, and it might be the least funny one we have. That is also why people keep pointing at it.
It is not a punchline about a bad deploy. It is not a 2 a.m. incident line, although it has saved a few nights. It is a reminder that judgement is still a job.
A model can summarise a thousand logs. It can draft the runbook, spot the pattern, and suggest the fix. What it cannot do is care about the consequence. It does not know which customer is affected, which alert is noise, or which quiet detail in a ticket changes the whole decision.
A human does. Or at least, a human is supposed to.
Three places the loop still matters
I keep coming back to three ordinary moments where this phrase shows up in real work.
First, the approval. Someone has to read the change, not just watch the green tick appear. Green means the checks passed. It does not mean the idea was good.
Second, the exception. Automation loves the normal case. Real life sends the weird case at 4:47 p.m. on a Friday: the account with the edge condition, the order with the odd address, the user who does not fit the flow. The loop is where that person gets seen instead of rejected by default.
Third, the aftermath. When something breaks, a tool can generate a timeline. It takes a person to say, honestly, what we missed, what we will change, and what we will not pretend was unforeseeable.
None of this is anti automation. I automate happily, and this shop only exists because print on demand automates the parts I could never do alone at a kitchen table. The point is narrower: automate the work, keep a person on the decision.
Why put it on a hoodie?
Because reminders work better when they are visible.
A policy document lives in a folder nobody opens. A phrase on a hoodie in a standup, a review, or a call tends to get read out loud. Once it is read out loud, it does its job. Someone pauses. Someone asks the extra question. Someone says, wait, did a person actually look at this?
That pause is the whole product, really. The cotton is just the delivery method.
It also works as a quiet signal between strangers. When another tech person sees the phrase, there is no explaining to do. They have been the human in somebody else's loop. They have also been grateful, at least once, that a human was in theirs.
What I am learning from this one
Of all the designs in the shop, this is the one people treat like a statement rather than a joke. That surprised me at first. I expected the DNS jokes to do all the talking.
It turns out there is room for both. Some days you want the line that makes the incident channel laugh. Some days you want the line that reminds the room to slow down before the incident channel is needed at all.
If you work with automated systems, I would love to hear your version of this. Where does your team insist on a human, no matter how good the tooling gets? Code review, releases, support, hiring, content, safety? And where have you happily handed the loop over, and never looked back?
Tell me in the comments. The honest edge cases are always more interesting than the policy.
COMPILE & COTTON >_ is my small shop of wearable inside jokes for people who build the internet: https://compileandcotton.etsy.com
I also write the Deploy Day Digest, a short read for tech folks. Join here: https://deploydaydigest.beehiiv.com/subscribe and use code DIGEST10 for 10% off in the shop.
Top comments (0)